New Reasoning AI Models Retain Medical Stereotypes

Researchers at Flinders University found that advanced reasoning AI models still produce clinical vignettes containing racial and gender stereotypes ^[1]. The study analyzed 36,000 cases generated by next-generation models, including o3-mini and DeepSeek-R1 ^[1][2]. Despite improved computational logic, these systems continue to replicate historical biases found in healthcare data ^[1].

The findings suggest that better reasoning capabilities do not automatically fix representational errors in medical applications ^[1][3]. Experts warned that these biases could impact patient care if integrated into clinical workflows without strict safeguards ^[2]. The research highlights a persistent gap between technical performance and social fairness in artificial intelligence ^[1][3].